> ## Documentation Index
> Fetch the complete documentation index at: https://docs.pureframe.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# MCP Server

> The @pureframeai/mcp package — install, run, and deploy options.

`@pureframeai/mcp` is the package that turns your Pureframe AI library into an MCP server. For per-client setup instructions (Claude, Cursor, VS Code, Codex, OpenCode), see [Connect an Agent](/agents/connect-an-agent) — this page covers the package itself.

## Running it locally

Most clients invoke this for you via `npx`, so there's usually nothing to install manually:

```bash theme={null}
npx -y @pureframeai/mcp
```

It reads `PUREFRAME_API_KEY` from the environment and speaks MCP over stdio — this is what every client config in [Connect an Agent](/agents/connect-an-agent) points at.

## Remote endpoint

Pureframe AI also hosts the same server as a Cloudflare Worker at `https://mcp.pureframe.ai`, speaking MCP over HTTP. Use this instead of the local `npx` process for clients that support remote MCP endpoints (Claude.ai web, for example) — no Node.js runtime required on your end. The trade-off: the remote endpoint can't read your local filesystem, so `upload_video` isn't available through it.

## Tools exposed

`search_videos`, `list_collections`, `get_collection`, `get_video`, `upload_video` (local only), `get_job_status` — see [MCP Overview](/agents/mcp-overview) for what each does.

## Source

The package wraps the same REST API as the [Python](/sdks/python-sdk) and [JavaScript](/sdks/javascript-sdk) SDKs — if you need a capability the MCP tools don't expose, call the REST API or an SDK directly alongside it.